Alfred A. Brooks 100 Wiltshire Drive Oak Ridge, TN 37830-4505 Phone (& fax/call): 423 482-1559 E-mail: brooks50@comcast.net Note: The following is an ASCII E-mail version of an MS WORD document. Some style changes have occurred. Two Superfund Models: Their Assumptions and Consequences ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Alfred A. Brooks - September 1996 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Abstract: The paper contrasts two cost models for the Superfund remediation program and concludes that, based upon optimization theory, the constraint of resources places distinctly different requirements on the risk information base than does the currently unconstrained resources model. It is further concluded that the current biased information base is counter-productive under the constrained model as it leads to greater overall residual risks rather than lower residual risks. The constrained model requires a new approach to the problem. Introduction ~~~~~~~~~~~~ The scope of this discussion is limited to the Comprehensive Environmental Response, Compensation and Liability Act but as others have pointed out the scope of Federal risk assessment[1] goes far beyond this undertaking. The total cost of federally required risk assessment is estimated[2] to be $600 billion per year when private sector mandates are included. The conclusions of this paper are also applicable to this broader scope. The Superfund Remediation Program as defined by CERCLA comprises a collection of contaminated sites detailed in the growing National Priorities List. The CERCLA legislation requires attention to "significant health risk" and "cost-effective remediation". The resources estimated for the remediation of all sites (and in some cases, for single sites) are large by any measure varying upward from $300 billion for approximately 2500 sites. These sites must compete for resources within the applicable agency and must compete with other tasks especially within the Federal budget. To the author's best knowledge, an underlying analytical cost model for Superfund, its assumptions and its consequences has not been systematically defined or discussed. This paper discusses in similar terms two such models varying by one assumption and contrasts their data requirements. The first model represents the Superfund program as it has been and is now conducted. The model is inferred from the on-going EPA operational methodology observed in the existing remediation program. The second model changes a tacit assumption of the first model, i.e., the availability of resources, and notes a rather surprising change in the character of the data required The models are developed only to the extent necessary to reveal this fundamental, important difference. The Current Superfund Methodology ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Currently the development of a Superfund site remediation plan consists of two steps: 1) a Risk Assessment phase during which an analysis of the risks is performed and a remediation plan is proposed by qualified experts, and 2) a Risk Management phase during which the public is given the opportunity to comment on the advisability of proceeding with the plan or possibly to modify it. The current risk assessment methodology is to establish on a site by site basis the following: 1. The contaminant(s) of concern for the site. 2. The exposure pathways and doses for humans and other eco-species. 3. The acceptable regulatory dosage levels for humans and impacts for the ecosystem. 4. Remediation level(s) which are protective of the relevant species. 5. Alternative remediation strategies which obtain the proposed remediation levels. The preferred methodology for the analysis of each site would be to construct an accurate, site-specific, risk distribution function and determine by a statistical methods what contaminant level would be protective of some acceptably high percentile of the exposed population. Unfortunately the set of risk distribution functions and other accurate data necessary to carry out such an analysis is not available and can be made available only by considerable effort. Very little human data is available and most of the risk data has to be inferred by extrapolation from animal studies. This extrapolation involves a great deal of uncertainty. In addition, there are uncertainties in the extrapolation from the high experimental doses to the low dose regions of actual exposure. In the face of such uncertainties and believing that it is better to spend more money than to jeopardize human or environmental health, the regulators have understandably chosen to use conservative approximations to the risk parameters in the belief they would commit an "always safe error". The detailed methodology is often determined by the EPA guidelines[3] or other statutory authority whose values are often very conservative so as to be applicable to a wide variety of sites without predicting remediation levels which would not be protective of health. Similar margins of safety are applied to the choice of exposure pathways and their parameters. The results are deliberately designed to be a conservative bound on a remediation value which would by itself be protective of a high percentile of a sensitive subpopulation of the exposed population as predicted by a complete statistical analysis of the risks. The repeated use of conservative approximations has led to many cases of very conservative, proposed remediation goals and, in the eyes of many critics, unnecessary remediation and hence wasted resources. The wide range of risk avoidance costs encountered can be seen in a report by Teng et al[4]. Many of the projected risks lie well below the ability of epidemiology to verify[5]. Some critics have gone so far as to label the risk assessment as bad science[6] or even non-science[7]. The conservative nature of risk assessment has attracted the attention of prominent editors[8] as well as Congress[9]. That this methodology leads to excessive remediation is borne out by the experience at the Lower East Fork Popular Creek Superfund Site where citizen pressure to revise the very conservative mercury risk data led to a change in proposed risk level[10] of 10 ppm Hg and a remediation cost of $1.6 billion to a final level[11] of 400 ppm and a cost of $8 million, a 200 fold reduction in cost. Many regulators recognize this conservatism and would seek to remedy it but find it difficult in the face of the great uncertainties that now prevail in the risk data base. It is not the purpose of this paper to debate the issue of whether this methodology has been effectively used but to address a more fundamental question: the cost model used and its implications for the database required. While the current methodology may consider the costs of alternative remediation methods in order to reduce the costs at one site, there is little effort made to consider cost reductions among the many sites by equating the real risks at the several sites. In other words, the sites are considered to be independent and, for all intents and purposes, resources are considered as unconstrained. The methodology also precludes the comparison to other risks which lie in the purview of other agencies and in budgets that are not directly competitive with the Superfund budget except in the purely political arena of Congress. There is also a tendency to strive for zero risk as the only acceptable risk and a "not in my backyard" and a "not on my watch" posture by residents and risk administrators. Whereas in reality risks are never zero for mortals and some level of risk must be accepted[12]. The real question is: Are we truly minimizing the residual risk with the resources at our disposal? The Unconstrained Resources Model ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The unconstrained cost model is quite trivial. The only reason to describe it is to examine its assumptions and consequences in order that they may be contrasted directly with another model. Under the current methodology described above, the total cost is simply the sum of the costs of the independent sites: n C[n] = SUM C[i](r[i](l[i])) i=1 where C[n] = the total cost of n sites C(i) = the cost function of the i-th site r(i)(l[i]) = the risk removed from the i-th site, and l(i) = the level of the contaminant(s) at the i-th site. The functional form of cost and risk do not preclude the consideration of future cost of both risk and resources or other complex risk functions. These do not change the additive nature of the cost model but only the difficulties in expressing its parameters. The assumption of site independence implies: 1. the r[i]s and l[i]s are independent and each may include its own independently determined conservative bias. 2. Increasing n will increase C[n], the total cost, by the additional cost of the new site(s). 3. Increasing C[i], the cost at any site, will likewise increase the total cost, C[i] but not the costs at other sites. There is no established limit to the number of conservative data bounds used to determine the bound on the proposed remediation goal. The cost of each site is determined in part by the degree of conservatism in the assumed risk bounds which in turn stems from the assumption of "always safe error". Operationally as sites are added to the National Priorities List or the cost at any site increases, the total cost will increase as needed. In this sense the cost of the Superfund program is unconstrained. Thus the concept of an "always safe error" is logically equivalent to "unconstrained resources". This model does lead to safe remediation but it does so by increasing costs while insuring the results at each site are always safer than required. For the purposes of this discussion, the important consequence of this model is that conservatively biased risk data is acceptable to its stated goals, i.e., unquestionably safe sites, and to the scientific methodology used to reach those objectives, i.e., always safe data bounds. The Changing Political Climate ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Up until 1994 the political climate encouraged the belief that resources could always be made available and, in that climate, the related assumptions of "site independence", "unconstrained resources' and "always safe error" may have seemed reasonable in spite of the protests alleging waste and non-science. At least it was a short-term, safe solution to the problem in the face of uncertain data. It is clear that the political events of 1994 with the demand for a balanced Federal budget have significantly changed the ground rules for major Federal projects including Superfund. The DOE alone has made massive cuts in the projected funding. They used to discuss "greenfields" and "residential criteria"; they now discuss "brownfields" and "abandoning in place". Major projects[13] and technology developments are faced with shortfalls in their budgets. It is time to face the question: What changes does the constraint of resources make upon the Superfund remediation program, its model and the pursuit of acceptable solutions? The Constrained Resources Model ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Constrained systems possess an optimum strategy which for the specific constraints obtains the maximum possible response. Responses exceeding this are obtained only by expending more of the constrained resources. That this optimal strategy can be difficult to discern is immaterial to its existence and to its influence on the problem being addressed. Constrained in this context means less than that necessary to obtain the maximum response. The collected Superfund sites can be remedied by many strategies each removing its associated risk. When remediation resources are limited one of these strategies removes the most risk for the limited resources available. In other words the risk response surface has a maximum which represents the most accomplishment; all other strategies leave more risk on the collected sites albeit lower at some sites. Operationally the parameters of the maximum determine how much of the resources are to be allotted to each site and how they are to be applied to that site. Further any other allocation is ensured of being a non-conservative error, i.e., less remediation than was possible. In other words; there is no "always safe error" such as existed in the unconstrained problem. The total cost for this model is: n C[n] = SUM C[i](r[i](l[i])) = a fixed quantity Eq. 1 i=1 and the collective risk removed is: n R = SUM r[i](l[i])) i=1 and the l[i]s are selected to maximize R and to simultaneously satisfy Eq. 1. The remaining symbols have their prior meanings. The subscripted functions and values collectively define the optimal strategy. This model has the following important characteristics: 1. The optimal strategy, inherent in constrained systems, exists whether or not an attempt is made to find it. 2. All other remediation strategies leave a greater residual risk. The model is somewhat analogous to being on a mountain range in the fog. There is a highest peak somewhere whether you can see it or not or whether you climb it or not. A highest peak is a characteristic of mountain ranges, not mountain climbers nor their intentions. The analogy can be carried further: if the optimum strategy for gaining the highest mountain peak requires a minimum of twenty days and you only have ten days, there is an optimum strategy which will gain the most altitude in the allotted time. Further if one attempts to ascertain an optimum strategy from maps and photographs, they should be accurate. The probability of selecting the optimum strategy for the Superfund program using erroneous data is vanishingly small when the number of sites is large or when the exposures are complex and involve many parameters. Clearly the risk assessment data base established under the assumptions of the unconstrained model is quite unsuitable for solving problems under the constrained model. The author has conducted simple optimization computations for the LEFPC Superfund Site[14] using a soil mercury distribution function supplied by the project. Data biases as little as a factor of five or ten reduced the total risk reductions by thirty percent and completely eliminated the remediation of one of three identical sites. These biases are less than the typical EPA safety factors. The inherent nature of the constrained system has denied us the luxury of the "always safe error". This is unfortunate but true. Unbiased data is of the utmost importance to the solution of these problems under the constrained model. What may be worse is that, while under the unconstrained model each departure from accurate data leads only to the needless expenditure of funds, under the constrained model inaccurate data leads to increased public or environmental risk. Thus the current data base is counter productive. Conclusions and Application of the Constrained Model ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ It is apparent that the two models may differ only slightly in assumptions but drastically in conclusions. Thus it is important that the correct model be adopted and it is difficult to defend the assumption of unconstrained spending in view of the recent political trends. A comparison of the two models leads to the following conclusions: 1. The unconstrained resources model obtained a degree of safety in remediation by adopting conservative bounds to the problems and by allotting more resources to independent sites thus contributing to the total allocation. 2. The constrained resources model disallows the allocation of additional total resources and improvements in safety must be brought about by more cost-effective allocations among sites within the overall resource constraints in a manner which will maximize the total hazard removed from the sites. 3. To obtain increasingly closer approximations to an optimal remediation strategy will require increasingly accurate, unbiased data to be used in unbiased risk models. 4. The requirements for data for the constrained model are distinctly different from those for the unconstrained model, i.e., unbiased and accurate vs. biased and approximate. This distinct difference requires a new approach to the establishment of the information base supporting the risk assessment requisite in the constrained model. In the best of worlds, one would have a complete list of sites, the needed information base, the detailed site models and proceed to optimize the problem subject to the resource constraints. Clearly we do not live in the best world: 1) the list of sites is a moving target, 2) the information base is absent, 3) the detailed models are poorly understood and 4) optimization using a probabilistic response function may be prohibitively costly. Nonetheless, Superfund risk assessment should not continue to be based on data not suitable for the model which is applicable and essentially different from the prior model and data. Instead we should make the best approximations to the required data and detailed models. Only by approximating the proper model can we approximate reality and avoid the pitfalls of using the wrong model. The primary pitfall of the unconstrained model is: When the constraint on the resources is reached the remediation process ceases in spite of the fact it is not yet complete. True there is no hard and fast limit to the allotted resources but there must come a point where the public and Congress, not convinced of real risks to be remediated, says "Enough is Enough". In the private, sector non-optimal allocations reduce our economic efficiency. The following steps outline the necessary approach to changing models: 1. Determine acceptable statistical models to be used and acceptable protected percentiles. 2. Determine a set of acceptable unbiased distribution functions. 3. Examine the current information base to remove all biases using the limited occupational data and accident data to place the best unbiased human and animal data into a new database suitable to the new models. 4. Carry out the necessary research to remove bias from the extrapolations of animal test data to the human species and to determine other risk parameters. In carrying out the above it is of utmost importance that the risk assessment and risk management community understand there no longer is a "safe haven" of "always safe error" and that this affects every decision and approximation in the risk assessment process. It is also important that the public have confidence in the risk predictions and support the elimination of the real risks they deem unacceptable. In closing, the regulatory community has resisted reducing the safety factors used in the regulatory process based on the belief that in the face of uncertainty that the safety factors led to the protection of the public. However in the face of constrained resources, there is every reason to believe that the opposite is true. This is ample reason to adopt a more appropriate and accurate risk information base. Granted we shall not reach the full optimal strategy but we should at least try to approximate it. _________________________ The author is a resident on the LEFPC Superfund site and a retired chemist. References: [1] J. D. Graham et al.; A Historical Perspective on Risk Assessment in the Federal Government, March 1994; Harvard Center for Risk Analysis, 718 Huntington Ave., Boston, MA 02115 [2] Reform of Risk Regulation: Achieving More Protection at Less Cost - Report of the Harvard Group on Risk Management Reform (Att'n: Dr. J.D Graham); March 1995; Harvard Center for Risk Analysis; 718 Huntington Ave.; Boston, MA 02115 [3] Risk Assessment Guidance for Superfund Volume 1 - Human Health Evaluation Manual ( Part A,); EPA Pub 540/-89.002, Dec 1989 Risk Assessment Guidance for Superfund Volume 1 - Human Health Evaluation Manual ( Part B, Development of Risk-based Preliminary Remediation Goals); EPA Pub 9285.7-01b, Dec 1991 [4] T. O. Teng; Five-Hundred Life-Saving Interventions and Their Cost-Effectiveness, July 1994 (Draft); Harvard Center for Risk Analysis, 718 Huntington Ave., Boston, MA 02115 [5] Taube, Gary; Special News Report - Epidemiology Faces Its Limits; Science, Vol 269, 1995/7/14, p.164 Seiler, Fritz, et al.; The Scientific Method in Risk Assessment; Technology: Journal of the Franklin Institute, Vol 331A, pp. 53-58, 1994 [6] CURE, Risk Assessment: Resolving the Controversy, Nov 1994; Coalition for Uniform Risk Evaluation, 1747 Pennsylvania Ave., Washington, DC 20006, Ph: 202 833-5055 [7] Gough, Michael; It's Not Science. What Can Science Do About It?; Health Physics, Vol. 71, No. 3, Sept. 1996, pp. 275-278 Seiler, Fritz, et al.; Definition of a Minimum Significant Risk; The Scientific Method in Risk Assessment; Technology: Journal of the Franklin Institute Vol 331A, pp. 83-95, 1994 [8] Abelson, Phillip; Editorial: Reflections on the Environment; Science, Vol 263 1994/2/4 p. 591 Abelson, Phillip; Editorial: Flaws in Risk Assessments; Science, Vol 270 1995/10/13 p. 215 [9] Breyer, Hon. Stephen, Testimony at Hearing on Use of Risk Analysis and Cost-Benifit Analysy in Setting Environmental Priorities, Comm on Energy & Natural Resources, 11/9/93 Johnson, Sen. J. Bennett, Congressional Record-Senate May 18, 1994 S5859-5909 [10] Bashor, B.S. and Turri, P.A.; A Method for Determining an Allowable Concentration of Mercury in Soil; Arch. Environ. Contam. Toxicol. 15, 435-438(1986) [11] Record of Decision for Lower East Fork Poplar Creek DOE/OR/02-1370&D1; May 1995 [12] Reilly, W. K.; Risky Business: Life, Death, Pollution and the Global Environment; 94/01/12; Institute for International Studies; Stanford University, 200 Ecina Hall, Stanford, CA 94305-6033 [13] The 1996 Baseline Environmental Management Report - Executive Summary; DOE/EM-0290; June 1996 Oak Ridge Operations Office Environmental Management Ten Year Plan, Oak Ridge, Tennessee(Draft); July 1996 [14] East Fork Poplar Creek - Sewer Line Beltway Remedial Investigation Report; DOE/ORO by SAIC; April 1993 and Addendum May 1994.